The false negative rate may be very low, this doesn't speak to the false postive rate. Considering the goal was supposed to be "telling humans and computers apart" both kinds of errors are important to consider. Google predicates their captcha primarily on the logged-in user's activity around the web, failing that they use the ip. A human being not logged in and using an open proxy is likely to get shut out completely. It ceases to be a captcha, in the traditional sense, at that point. This may be a good thing, depending on your standpoint.